Tommaso, As it sounds like you've seen, there is a Project Clearwater capacity spreadsheet at http://www.projectclearwater.org/technical/clearwater-performance/. The numbers in this spreadsheet are based on the results of stress testing using a standard SIPp script (https://github.com/Metaswitch/sprout/blob/icscf/tests/load/call_load2.xml),where the script simulates a pair of subscribers registering every 5 minutes and making a call to each other every 30 minutes. The script runs on Clearwater stress nodes against the bono cluster in a deployment, each node accounts for 100,000 subscribers, and multiple stress nodes can be set up. This testing is done on a complete deployment then, rather than measuring individual components. The per-node results are calculated from a system running 3 million BHCA under a balanced load, and noting how many of each node are needed. We don't currently have any results from the stress testing we've done with SIPp that allow a comparison of different user profiles.
It's worth bearing in mind that the values in the spreadsheet are actually a little out of date now, as we've been making substantial changes to Clearwater recently, including adding additional function, and expect to continue for the next few sprints. Once these changes are in we expect to refresh the performance numbers. Information on the stress testing, and details of how to set it up if you want to get some values out yourself, potentially modifying the tests to cover the profiles you're interested in, can be found at: https://github.com/Metaswitch/clearwater-docs/wiki/Clearwater-stress-testing. In addition to the existing performance numbers and SIPp testing, we've also recently done some benchmark testing with IMSBench. A Clearwater deployment consisting of 11 m1.small nodes (3 bonos, 2 sprouts, 4 homesteads and 2 homers) was set up, and 100k subscribers were provisioned on the deployment. The scenarios attempted in the benchmarks tests split into 50% calls, 30% SIP MESSAGEs, 15% re-registrations, 2.5% new client registrations and 2.5% de-registrations. The IMSBench tests ran successfully up to 340 scenario attempts per second. As 17.5% of the scenarios were registrations or re-registrations, this translates into ~60 re-/registrations per second on the deployment, while it was also running 1.2M BHCA. I hope this helps and let me know if you have any other questions. Ellie -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Tommaso Cucinotta Sent: 09 December 2013 17:48 To: [email protected] Subject: [Clearwater] Performance of ClearWater @ AWS Hi, I'd like to ask whether there's any standard measurement of how ClearWater performs when installed on Amazon instances. I've seen a spreadsheet seemingly containing numbers referring to individual components measured/profiled autonomously, then some calculations / deductions made based on those figures. Instead, I'd like to know what numbers to expect (e.g., number of authenticated registrations per second) from a complete end-to-end interaction through bono, sprout, homestead and back to the user with, let's say, an installation with 1 (m1.small) VM per component. Can anyone provide such experimental figures? Thanks, T. -- Tommaso Cucinotta, Computer Engineer PhD Researcher at Alcatel-Lucent Bell Laboratories Blanchardstown Business & Technology Park Dublin - Ireland _______________________________________________ Clearwater mailing list [email protected] http://lists.projectclearwater.org/listinfo/clearwater _______________________________________________ Clearwater mailing list [email protected] http://lists.projectclearwater.org/listinfo/clearwater
